Detecting Temporal Anomalies in Business Processes Using Distance-Based Methods

Abstract

Outlier detection in process mining refers to either infrequent behavior in relation to the underlying business process models or to anomalous latencies of task execution (temporal anomalies). In this work, we focus on the latter form of anomalies and we propose distance-based methods. Compared to solutions relying on probability distribution analysis and based on the experimental evaluation presented, our proposal is shown to be capable of covering both trace and event outliers, and being more efficient and effective. More specifically, running times of our technique are lower by up to an order of magnitude, while we achieve significantly higher precision and recall.